More people are using AI than ever, yet the stocks of companies building AI infrastructure are falling. Management's assurance that demand is strong is no longer enough to hold up share prices.
According to a Reuters report dated October 15, 2026, the Philadelphia Semiconductor Index had fallen 11.6% over the six trading days up to that point. At the center of the slide were concerns about the financing of CoreWeave, a company that provides computing services built on Nvidia GPUs. Jensen Huang's response, though, was strikingly different from the market's. He struck an optimistic tone, saying CoreWeave's financial health was improving. [1][2]
The person closest to the AI industry is optimistic—so why is the market nervous?
Perhaps the two sides are answering different questions. Huang is looking at how much computing power will be needed going forward; investors are looking at when the money borrowed today to supply that computing power can be recovered.
The Company That Sells GPUs vs. the Company That Makes Money With Them
CoreWeave rents out GPU computing capacity to companies that develop and run AI models. Customers get the computing power they need without building data centers of their own. CoreWeave, in turn, must pour big money into equipment and infrastructure upfront to deliver that service. Reuters identified this business model—heavy upfront investment funded by debt—as the crux of the current unease. [1]
This is where the difference between chip suppliers and infrastructure operators shows.
From an investor's standpoint, revenue from selling equipment must be distinguished from revenue earned by running that equipment for years to recoup the investment. Even under the same premise of strong AI demand, the risks you need to verify change depending on which side you've invested in.
There can be a longer distance than expected between "GPUs are selling well" and "the shareholders of the company buying those GPUs earn good returns."
Revenue Doubled, Yet Questions Remain
It is hard to dispute CoreWeave's growth itself.
In its second-quarter 2026 results announced on August 11, revenue came in at $2.575 billion—up roughly 112.5% from $1.212 billion a year earlier. Yet in the same quarter, the operating loss was $49 million, net interest expense was $640 million, and net loss was $626 million. Net interest expense alone came to about 24.9% of revenue. The ratio is calculated from the company's reported figures. [3]
These are numbers where rapid revenue growth and a heavy financing burden both hold true at once.
These results alone cannot seal a verdict that the business is failing. For a company shouldering costs upfront during an expansion phase, current profit and loss alone may be a poor guide to the future. Still, as an investor, one more question should follow the revenue growth rate.
As the business scales, will the money left over for shareholders grow with it?
The company also disclosed a revenue backlog of roughly $104 billion as of the end of June. That is a significant amount. Under the company's definition, however, the figure includes remaining performance obligations and amounts it estimates will be recognized as revenue in the future under contracts, and it is contingent on meeting service delivery and availability requirements. [3]
As an investor's reading, backlog is evidence of visibility into future business—not cash sitting in the bank today. What matters is not only the size of the contract amounts, but the sequence of the money spent first to realize that revenue and the cash actually collected.
Jensen Huang's Optimism Has Both Grounds and Self-Interest
That said, it is also hard to write off Huang's remarks as mere words to soothe a customer.
When Nvidia announced an expanded partnership with CoreWeave on January 26, 2026, it disclosed that it had invested $2 billion at $87.20 per share. The two companies also laid out a plan to accelerate the buildout of more than 5 gigawatts of AI infrastructure by 2030. Huang didn't just talk—he put capital in as well. [4]
At the same time, this relationship is a backdrop that cannot be overlooked when reading Huang's remarks. Because Nvidia has invested directly in CoreWeave's growth, his optimism is a judgment on the industry and at the same time a judgment on his own investment. There is no need to ignore this—but there is no need to take it as the assessment of a disinterested third party either.
The ties between the two companies do not stop at the equity stake. According to CoreWeave's September 2025 contract filing, Nvidia is obligated to purchase remaining unsold computing capacity under the agreement through April 13, 2032, subject to conditions such as termination clauses and service delivery and availability requirements. [5]
My interpretation is that this is a mechanism to cushion demand risk. However, a purchase commitment covering a specific amount of capacity must not be read as a guarantee of all the company's obligations or of its share price. Nvidia's involvement and the claim that CoreWeave shareholders' investment risk has disappeared are two different stories.
What the Market Is Watching May Be the "Timetable of Money," Not Demand
On October 15, Huang said, in effect, that CoreWeave's cost of capital was coming down. In the bond market, meanwhile, it was reported that yields on some CoreWeave bonds had climbed to around 15%. Here, the funding picture management conveys appears to diverge from the yields bond investors are demanding in the secondary market. [2][6]
This does not mean the rise in bond yields should be read as the interest rate on all existing borrowings immediately moving to the same level. Still, it is worth examining what burden this could place on new financing and refinancing.
My own view of this selloff sits right here.
The market is not necessarily saying "AI is useless." It may be asking: "Spending on AI is real, but is the revenue to support that spending coming in fast enough?"
One can imagine a path in which harder financing leads infrastructure operators to slow investment, and the impact feeds through to equipment orders. This is not an already-decided ending but a downside scenario hardware investors should check. Conversely, if customer cash payments and facility operations continue as planned, today's unease may turn out to have been excessive.
References
Conclusion
I do not want to rush to boil this plunge down to either "the collapse of the AI bubble" or "a chance to buy a good stock cheap."
Jensen Huang's optimism and the market's anxiety can coexist. Even if more computing power will be needed going forward, not every company supplying that computing power becomes a good investment.
I believe there is now one more question to ask in AI hardware investing.
Alongside "Who makes the best equipment?", we must also ask: "Can the customers buying that equipment hold out until they recoup their investment?"
To me, this CoreWeave episode looks less like the end of AI demand and more like the moment investors start evaluating suppliers and infrastructure operators—the two long lumped together in one rally—as separate businesses. The next gap in investment results may come not from conviction in the technology, but from how closely one examines the process of turning that conviction into cash flow.
Disclaimer: This article is informational content summarizing public filings, analysis, and the author's views; it is not a recommendation to buy or sell any particular stock.